Scene Parsing with Global Context Embedding

This repo is the caffe implementation of the following paper:

Scene Parsing with Global Context Embedding
Wei-Chih Hung, Yi-Hsuan Tsai, Xiaohui Shen, Zhe Lin, Kalyan Sunkavalli, Xin Lu, and Ming-Hsuan Yang. In ICCV, 2017.

Please cite our paper if you find it useful for your research.

@inproceedings{hung2017scene,
  title={Scene Parsing With Global Context Embedding},
  author={Hung, Wei-Chih and Tsai, Yi-Hsuan and Shen, Xiaohui and Lin, Zhe and Sunkavalli, Kalyan and Lu, Xin and Yang, Ming-Hsuan},
  booktitle={IEEE International Conference on Computer Vision (ICCV)},
  year={2017}
}

Prerequisite

• DeepLab-v2 caffe. You will need the this updated version [link] for most recent machine setups.
• A GPU with at least 12GB

Test on ADE20k validation set

• Download the ADE20k dataset and put it in data/.

The directories should be like this:

./data/ADE20k/annotations/validation
             /images/validation

• Download pretrained model

bash get_model.sh

• Download precomputed context features/priors of ADE20k val set.

bash get_prior.sh

• Execute evaluation script:

python eval.py --prototxt prototxt/ade20k_val.prototxt \
               --model models/ade20k_full.caffemodel \
               --save-dir results/ade20k/val/ \
               --gpu 0

The result images will be saved at results/ade20k/val/.